AI Insights and Trends Summary — 2026-09-23

Key Technical News and Announcements

• SoftBank issues over $11B in high-yield bonds to secure OpenAI investment funds — Brussels is preparing A-G energy rating systems for data centers, and a UN science panel is also discussing AI-related issues, showing that the AI boom is expanding beyond product launches into capital and regulatory environments.

• Salesforce unveils AI agent platform "AIForce" at Dreamforce 2026 — A platform where AI directly operates software to handle tasks, combining corporate data and business rules with AI to produce consistent results while integrating with Claude and Slack.

• "Scientific AI" expands into scientific research — The scope of AI is rapidly expanding beyond text and code generation into scientific research. Recently, moving past paper searches and knowledge summaries, cases where AI reasons through complex scientific problems and directly participates in the research process are on the rise.

• Anthropic key developments — Alongside the issue of Claude Opus 5 "penetrating" OpenAI, regulatory news such as Trump's mention of AI safety and Governor Newsom's kill switch executive order became major talking points.

New AI Products and Services

• Upstage Solar Mini 4 — Released on September 22, 2026. It supports a 512K context with a 3B active model out of 35B parameters, priced at $0.10/$0.40. It also comes with "Solar Jev," a decision-making endpoint that doesn't generate prose.

• Aion 3.5 Mini — Released on September 23, 2026. The context doubled to 262,144 tokens without a price change, and its base model transitioned from DeepSeek to the GLM family.

• YBM In-gang "AI Level Test" — An adaptive English test introduced by YBM Net where question difficulty adjusts in real time based on the learner's correct answers in about 20 minutes. It consists of 23 total questions covering vocabulary, listening, writing, reading, and speaking.

• Anthropic, Claude for Financial Advisors — Released for the Enterprise plan on September 15, providing 11 connectors including BlackRock, Schwab, and Vanguard. Additionally, on September 18, it named Accenture's Faculty division as its first embedded evaluator.

AI Learning and Educational Materials

• "Ability to think with AI" is the core of second-half AI education — In 2026, countries worldwide recognize AI not just as a tool but as a basic literacy, emphasizing that students must acquire competencies beyond mere AI utilization skills.

• Government AI education support — Expansion of computer science class hours and AI-focused schools — Summarizes policy trends where the legal status of AI digital textbooks was defined in August 2025 as educational materials in the form of "learning support software utilizing intelligence and information technology."

• Policy trends on strengthening basic AI literacy — Government policy stances driven by accelerating digital transformation are aligned toward strengthening basic AI literacy, expanding curriculum integration, and focusing on practical utilization-centered learning.

Analysis and Trend Insights

• Model competition: Narrowing gaps and China's quantitative lead — According to the Stanford 2026 AI Index Report, the gap among top-tier models has shrunk to just 2.7%. While the US still leads in top-tier AI models and influential patent production, China leads in paper, citation, and patent outputs, as well as industrial robot installations.

• Enterprise AI adoption is in the "workflow integration learning stage" — Analysis of over 17 million messages across 1,500 organizations shows that corporate ChatGPT output tokens increased about 7-fold in less than a year, yet companies are still actively learning how to integrate AI into their organizational workflows. Frontier companies (top 10%) show more than a 6-fold gap in skill usage rates compared to average companies (19% vs 3%).

• Robotics and physical AI are taking off — An IBM expert analyzed that "robotics and physical AI will certainly accelerate," noting that while LLMs remain dominant, the industry is reaching diminishing returns from scaling.

• MIT SMR's 5 AI Trends for 2026 — Davenport and Bean highlighted trends to watch in 2026, including AI bubble deflation and economic impacts, and the growth of "factory" infrastructure for companies adopting all-in AI.

Some screenshot-based information may be incompletely extracted, so please verify important details directly on the original source pages.